NMC Scientist Develops AI Model to Generate Realistic 3D Trees for Wildfire and Forest Research
NMC Scientist Develops AI Model to Generate Realistic 3D Trees for Wildfire and Forest Research
New Mexico Consortium (NMC) scientist Anthony Marcozzi is the lead author of a newly published study introducing an artificial intelligence model that can generate highly realistic three-dimensional representations of individual trees using only a small set of common forest inventory measurements.
The paper, “TreeFlow: Conditional Flow Matching for 3D Tree Point Cloud Generation from Inventory Attributes,” was recently published in the journal Multidisciplinary Digital Publishing Institute (MDPI).
Three-dimensional tree models are becoming increasingly important for applications ranging from wildfire behavior modeling and forest management to ecological research and digital twins of forest ecosystems. While modern laser scanning technologies can capture detailed tree structure, collecting these data across large landscapes is often expensive, time-consuming, or simply impractical. As a result, many existing models rely on simplified geometric shapes that fail to capture the complex branching patterns, canopy gaps, and natural variability found in real forests.
To address this challenge, Marcozzi developed TreeFlow, a generative artificial intelligence model that creates realistic 3D tree point clouds using readily available inventory information such as tree species, height, and data acquisition platform. The model was trained entirely on real terrestrial laser scanning data from the FOR-species20K benchmark dataset, allowing it to learn the complex structural characteristics of different tree species without relying on hand-crafted geometric rules.
The research demonstrated that TreeFlow produces synthetic trees whose structural characteristics closely match those of real scanned trees across multiple evaluation metrics. The model performed particularly well for trees under 25 meters in height, generating detailed representations that closely resemble real-world forest structure.
“Modern forestry and wildfire technologies increasingly depend on three-dimensional structure of canopies,” said Marcozzi. “However, this information is rarely available at the tree level and is often estimated with high uncertainty. This new modeling approach leverages advancements in generative AI to create hyper-realistic renderings of individual trees in 3D for use in fire simulations, wildfire risk assessment, and many other applications.”
By making it possible to generate realistic 3D trees from standard forest inventory data, TreeFlow has the potential to improve wildfire simulations, radiative transfer modeling, synthetic dataset generation, and digital representations of forests. The work also demonstrates how emerging generative AI techniques can help bridge critical data gaps in environmental science.
The publication represents another step forward in the NMC’s ongoing efforts to apply artificial intelligence and advanced computing to challenges in forestry, wildfire science, and ecosystem research.
To learn more, read the entire publication at:
Marcozzi, A., Tree-Flow: Conditional Flow Matching for 3D Tree Point Cloud Generation from Inventory Attributes. Multidisciplinary Digital Publishing Institute (MDPI).